A Neural Approach to Cursive Handwritten Character Recognition Using Features Extracted from Binarization Technique - Complex System Modelling and Control Through Intelligent Soft Computations - page 748

8.3 Experimental Conditions

There is not any rule that gives the calculation for the ideal parameter setting for a
neural network. However, the various parameters and their respective values used
in the proposed training process of the handwritten character recognition experi-
ments are shown in Table 2 .

9 Implementation and Functional Details

In the current situation, the number of neurons in the input and output layers are

fixed at 180 and 26 respectively. The 180 input neurons are equivalent to the input
character ' s size as we have resized every character into a binary matrix of size

15 1. The number of neurons in the

output layer is 26 because there are 26 English alphabets. The number of neurons in
the hidden layer and the activation functions of the neurons in the hidden and output
layers are to be decided.

12 and then reshaped to a matrix of size 180

× ×

Table 2 Experimental conditions during the recognition experiment
Parameters Value

Input layer
No. of input neurons 180

Transfer/activation function Linear

Hidden layer
No. of hidden neurons 80

Transfer/activation function TanSig

Learning rule Momentum

Output layer
No. of output neurons 26

Transfer/activation function TanSig

Learning rule Momentum

Learning constant 0.01

Acceptable error level (MSE) 0.001

Momentum term ( ) 0.90

α

Maximum epochs 100,000

Termination conditions Based on minimum mean square error or

maximum number of epochs allowed

Initial weights and biased term values Randomly generated values between 0 and 1

Number of hidden layers 1